Race models and analogy theories: A dead heat? Reply to Seidenberg

In two recent papers Seidenberg and his colleagues (Seidenberg, 1985; Seidenberg, Waters, Barnes and Tanenhaus, 1984) have argued that the regularity of the spelling-to-sound correspondence influences pronunciation of low frequency words but has no effect on pronunciation of high frequency words. Seidenberg (1985) also presents data which suggest that a similar effect holds for Chinese as well as for English. On this basis he concludes that skilled reading is unaffected by the manner in which the orthography encodes phonology. Seidenberg claims that even in different orthographies, phonology only influences the recognition of low frequency words. To account for his results Seidenberg has proposed a parallel interactive activation model of word recognition based on McClelland and Rumelhart (1981). Seidenberg argues that such a model is preferable to both a race model (dual code theory) of word naming and to analogy theories. In the present paper we will argue that Seidenberg’s claim that effects of spelling-to-sound regularity are limited to low frequency words in English is not adequately supported by the data. All crucial analyses in both Seidenberg papers are based on small numbers of items and do not generalise to new materials (see Appendix). Second, studies of phonologically shallow orthographies (Katz and Feldman, 1983) show that the manner in which the orthography encodes phonology does influence the reading process. Finally we will show that all of Seidenberg’s data can be equally well explained by the models which he rejects. Both a race model and an analogy theory can explain Seidenberg’s data on orthographic and spelling-to-sound regularity effects